Identifying Source Hotspots of “Non-Flushables” in Sewer Systems Through Machine Learning and Imaging Sensors

This thesis examines the feasibility of installing imaging sensors in sewers, combined with innovative machine learning techniques, to detect and identify non-flushable consumer products in sewers. A Raspberry Pi microprocessor with an off-the-shelf camera module was used, and Edge Impulse was applied to process captured imagery. The results indicated that optimal placement of the system (camera and lights) can vary depending on whether the products of interest float near the surface of the water or more towards the deep end of the sewer maintenance holes. The application of such a system for urban wastewater collection systems will be to proactively detect the areal hotspots of rising influxes of non-flushable consumer products (e.g., wet wipes and tissues). Building on knowledge gained on the performance and functionality of such a monitoring tool, utility managers can perform targeted public outreach, as opposed to general calls to the public to stop flushing consumer products.